It's worth watching the details when an agentic coder spins it's magic. You begin to see all the terminal commands it executes, a large number of them being greps and finds with globs, across your codebase. It begins jumping between files and reading different chunks here and there. Then as it codes you realise having one large file rather than many small components or screen-sized files of codes is preferrable for the agent. The agent can do something that we can't long pages of code.
As developers we preferrer shorter pages of codes, it suits our reading ability, our mental organisation of code into folder structures and so on. Also it helps avoid issues we have in collaborating on code across time, where several developers can work on different aspects of the code without conflicting with other developers working on different parts of the code.
However an agent operates very differently. They can read long pages, with ease, the generate code so fast and often work on their own. Based on these we could conclude, that agents are not pragmatic programmers, and forcing them to be, may not produce the best results. However allowing them to do good old practices, like long pages of html, or 'One Big Table' style databases, works to their strengths in fact.
In terms of what to do with the technical debt build up that results ... well that's a story for another post.
